Can You Gig It?

An Empirical Examination of the Gig-Economy and Entrepreneurship

Gord Burtch Seth Carnahan

Brad N Greenwood

UMN Symposium on the Sharing Economy May 2016

What I Hope You Remember

We investigate how the entry of gig-economy platforms (Uber X and Postmates) influences entrepreneurial activity in local areas.

Results suggest Marked decline in entrepreneurial activity after

platform entry Notably true for “low quality” entrepreneurs

Economic translation of implications Reduction of 1450 campaigns during the sample

$7.5 mm in requests over a 21 month period 14% decrease in requests one year out

Significantly reduces load on the platform by siphoning off low quality entrepreneurs

Platforms in the digital age

What do we know about platforms and the gig-economy?

Roughly $26 billion market (Malhotra and Van Alstyne 2014)

Research has progressed on many fronts: Platform design (Fradkin 2013, Fradkin et al. 2014)

Broader societal effects (Edelman and Luca 2014, Greenwood and Wattal 2015)

Effect on incumbent business models (Seamans and Zhu 2013; Zervas et al. 2015)

Most research suggests that the service economy is price sensitive and displacing lower tier competitors Hotels (Zervas et al, 2015)

Price sensitivity in drunk driving (Greenwood and Wattal, 2015)

But what do we not know? Most research has focused on the demand side of

these economies Limited focus on the supply side:

Racial bias and AirBNB (Edelman and Luca, 2014)

Peer to peer lending markets (Morse, 2015)

We are going to investigate who is likely working on these platforms, and how they are affecting local labor dynamics.

Research Question What is the effect of gig-economy platform

introduction on the rate and characteristics of entrepreneurial activity in a given locale? When we say entrepreneurial activity, we do not

mean working on the platform.

Why might these platforms influence entrepreneurial activity?

Two opposing forces may be at play: Existence of slack resources (Agrawal et al. 2015, Douglas and Shepherd 2000,

Greve 2007, Kerr et al. 2014, Richtnér et al. 2014, Shah and Tripsas 2007)

Opportunity costs (Acs and Armington 2006, Armington and Acs 2002, Åstebro et al. 2011, Block and Koellinger 2009, Fairlie 2002, Storey 1991)

These forces create a natural tension

Why might entrepreneurial activity rise?

Entrepreneurial activity is dependent on the existence of resources which can be re-assigned the venture (Agrawal et al. 2015, Richtnér et al. 2014)

Opportunities may be explored when resources are not constrained (George 2005, Shah and Tripsas 2007, Voss et al. 2008)

This also allows for low-cost, strategic sampling on the part of the entrepreneur

Entrepreneurship spikes when prestigious universities are on break (Agrawal et al. 2015)

Because gig-economy platforms allow the entrepreneur to set her own hours (Hall and Krueger 2015), she may devote resources to entrepreneurial activity without losing financial security

Why might entrepreneurial activity fall?

Many entrepreneurs have low opportunity costs (Åstebro et al. 2011, Block and Koellinger 2009)

These people pursue entrepreneurship as a means of resolving un- or under-employment (Block and Koellinger 2009, Fairlie 2002, Storey 1991)

Why? Entrepreneurial activity may have a higher expected

value than wage employment opportunities Or, the entrepreneur may be exploiting excess time

The entrance of gig-economy platforms may change the internal calculus for would-be entrepreneurs by offering a higher than otherwise expected wage.

So what is going to happen? If would be

entrepreneurs can re-deploy resources as a result of the inherent flexibility the platform offers, entrepreneurial activity may rise.

If the platform offers stable economic opportunities which allows entrepreneurs to capitalize on low opportunity costs, entrepreneurial activity may fall.

Context, Data, and Methods

The Setting Two Natural Experiments

The entry of the ridesharing platform Uber X The entry of the on-demand courier service

Postmates We first focus on Uber X We then extend the investigation with Postmates Experimental Validity

Multi-site entry which is geographically and temporally dispersed

Data Entrepreneurial Activity - Kickstarter

World’s largest crowdfunding platform (Burtch et al. 2013, 2015)

Significantly more reactive to changes Observation – EA/Month

3612 observations 21-month period between 2013 and 2015 172 Economic Activity

Areas Data on platform

entry is retrieved from: blog.uber.com Postmates.com

Variable Definitions and Estimator Model – Multi-site Diff in Diff Relative Time Model

𝑦𝑖𝑖 = 𝛼𝑖 + 𝜏𝑖 + �𝛽𝑗(𝑢𝑖∗ φ) + 𝜖𝑖𝑖𝑗

Dependent Variable (Two Forms): 𝑦𝑗𝑖 - ln(Number of Campaigns + 1) [OLS] 𝑦𝑗𝑖 - Number of Campaigns [Poisson]

Independent Variables: 𝜏𝑖 - Time Dummies (Quarters / Year and Seasonal) 𝑢𝑖 - Vector indicating if Uber X will ever enter EA i φ - Vector of relative time dummies 𝛼𝑖 - Vector of EA fixed effects

The objective is to measure the change in effect over time pre-

and post-treatment

(1) (2) (3) (4) DV ln(Campaigns) ln(Campaigns) Campaigns Campaigns Rel Time (t-6) 0.233*** 0.220*** 0.244*** 0.221***

(0.0764) (0.0502) (0.0581) (0.0550) Rel Time (t-5) -0.144 -0.0956 -0.0122 -0.0409

(0.0953) (0.100) (0.0709) (0.0643) Rel Time (t-4) -0.0299 0.0392 -0.0447 0.00392

(0.0488) (0.0476) (0.0520) (0.0579) Rel Time (t-3) 0.0508 0.0349 0.0477 0.0147

(0.0407) (0.0389) (0.0551) (0.0519) Rel Time (t-2) 0.0154 0.00262 0.0522 0.0173

(0.0344) (0.0333) (0.0376) (0.0414) Rel Time (t-1) Omitted Rel Time (t0) 0.0354 0.00182 0.0394 -0.00524

(0.0363) (0.0345) (0.0302) (0.0263) Rel Time (t+1) 0.0499* 0.0633** 0.0352 -0.00369

(0.0289) (0.0294) (0.0342) (0.0355) Rel Time (t+2) 0.0466 0.0568 0.00376 -0.0358

(0.0378) (0.0362) (0.0432) (0.0489) Rel Time (t+3) 0.0306 0.0210 -0.0257 -0.0818

(0.0451) (0.0412) (0.0571) (0.0673) Rel Time (t+4) -0.0234 0.00990 -0.143** -0.168***

(0.0449) (0.0441) (0.0581) (0.0648) Rel Time (t+5) -0.0314 -0.0249 -0.120* -0.150**

(0.0460) (0.0462) (0.0642) (0.0645) Rel Time (t+6) -0.0464 -0.0128 -0.196** -0.216**

(0.0590) (0.0657) (0.0764) (0.0860) Rel Time (t+7) -0.0938 -0.0838 -0.189** -0.222**

(0.0746) (0.0740) (0.0841) (0.0879) Rel Time (t+8) -0.160** -0.219*** -0.315*** -0.398***

(0.0648) (0.0641) (0.0773) (0.0787) Rel Time (t+9) -0.191*** -0.0935 -0.359*** -0.330***

(0.0678) (0.0703) (0.0881) (0.0949) Rel Time (t+10) -0.356*** -0.364*** -0.408*** -0.450*** (0.0782) (0.0784) (0.0965) (0.0990) Year Fixed Effects Yes No Yes No Seasonal Effects Yes No Yes No Quarter Effects No Yes No Yes N 3,612 3,612 3,612 3,612 R Squared 0 154 0 171

Robustness Checks

Selection Model The absence of a pre-treatment validates the

parallel trends assumption Employment related factors, that are time-varying,

may influence Uber’s decision to enter the market They may also influence entrepreneurship

Include Controls to control for employment dynamics: Log number of employed people Average Weekly Wage Total quarterly wages

(1) (2)

DV Campaigns Campaigns Rel Time (t-6) 0.238*** 0.219***

(0.0615) (0.0568) Rel Time (t-5) -0.0233 -0.0455

(0.0685) (0.0644) Rel Time (t-4) -0.0543 -0.000623

(0.0529) (0.0592) Rel Time (t-3) 0.0421 0.0128

(0.0567) (0.0528) Rel Time (t-2) 0.0498 0.0167

(0.0396) (0.0422) Rel Time (t-1) Omitted Rel Time (t0) 0.0311 -0.00832

(0.0308) (0.0283) Rel Time (t+1) 0.0249 -0.00759

(0.0333) (0.0360) Rel Time (t+2) -0.00505 -0.0389

(0.0446) (0.0501) Rel Time (t+3) -0.0350 -0.0849

(0.0592) (0.0686) Rel Time (t+4) -0.162** -0.175**

(0.0654) (0.0707) Rel Time (t+5) -0.150** -0.162** (0.0732) (0.0714)

Employment Controls Yes Yes

Year Fixed Effects Yes No Seasonal Effects Yes No Quarter Effects No Yes N 3,612 3,612

Number of Groups 172 172

Coarsened Exact Match Is there significant heterogeneity between treated

and untreated groups Important to minimize these differences

Coarsened Exact Match (CEM) (Blackwell et al. 2009, Iacus et al. 2009, 2012)

Population - to account for market size Average weekly wage - to account for the

differences in average local opportunity costs Current period

(1) (2)

DV Campaigns Campaigns Rel Time (t-6) 0.199*** 0.231***

(0.0752) (0.0839) Rel Time (t-5) -0.0503 -0.0386

(0.0961) (0.106) Rel Time (t-4) -0.0877 -0.00782

(0.0730) (0.0870) Rel Time (t-3) 0.110 0.0442

(0.0753) (0.0775) Rel Time (t-2) 0.0118 -0.00894

(0.0392) (0.0465) Rel Time (t-1) Omitted Rel Time (t0) -0.000814 -0.0374

(0.0298) (0.0294) Rel Time (t+1) -0.000433 -0.0482

(0.0397) (0.0469) Rel Time (t+2) -0.0113 -0.0651

(0.0406) (0.0505) Rel Time (t+3) -0.0481 -0.122*

(0.0509) (0.0670) Rel Time (t+4) -0.129** -0.192***

(0.0547) (0.0632) Rel Time (t+5) -0.139** -0.205*** (0.0608) (0.0704)

Year Fixed Effects Yes No

Seasonal Effects Yes No Quarter Effects No Yes N 2,895 2,895

Number of Groups 170 170

Diagnosing Standard Errors Serial correlation is a consistent concern with DD

estimations (Bertrand et al. 2002)

Inflates the probability of finding a significant result Random implementation Test

Randomly treat 1440 observations Replicate the estimation and store the coefficient Replicate 1,000 times

Benefits

Assesses the probability of spurious results (Bertrand et al 2002)

Reliable check against outliers

Random Implementation Test

Estimation Campaigns with Seasonal and Year

Fixed Effects Campaigns with

Quarter Fixed Effects μ of Random β -0.00007 0.00006 σ Random β 0.03482 0.03443 Estimated β (Rel Time t-4) -0.143 -0.168 Replications 1000 1000 Z-Score -4.105291 -4.881488 P-Value p<0.001 p<0.001

Postmates The results for Uber X are compelling

Startup costs for Uber X are still non-trivial Replicate these results with another platform

Postmates – on demand courier service

Benefits Replication to rule out spurious correlation / scientific

apophenia (Goldfarb and King, 2016)

If opportunity costs are the driving factor, a larger effect should manifest for a lower cost platform Postmates requires a bicycle Uber X a car in good condition

Rule out the argument that entrepreneurs are substituting the gig-economy platform for Kickstarter

(1) (2)

DV Campaigns Campaigns Rel Time (t-6) 0.0784 0.0795

(0.0637) (0.0545) Rel Time (t-5) 0.000571 0.00645

(0.0529) (0.0452) Rel Time (t-4) 0.0155 0.0459

(0.0364) (0.0362) Rel Time (t-3) 0.0311 0.0507*

(0.0313) (0.0277) Rel Time (t-2) 0.0532 0.0618**

(0.0366) (0.0248) Rel Time (t-1) Omitted Rel Time (t0) -0.0118 0.00244

(0.0344) (0.0464) Rel Time (t+1) -0.145*** -0.125**

(0.0411) (0.0567) Rel Time (t+2) -0.0965 -0.0937**

(0.0636) (0.0399) Rel Time (t+3) -0.171*** -0.160***

(0.0340) (0.0415) Rel Time (t+4) -0.112*** -0.0997**

(0.0408) (0.0455) Rel Time (t+5) -0.345*** -0.385*** (0.0706) (0.0724)

Year Fixed Effects Yes No

Seasonal Effects Yes No Quarter Effects No Yes N 3,612 3,612

Number of Groups 172 172

Significantly larger than Uber X Confirmed with pairwise Wald Tests

(2.35 and 2.42; p<0.05)

Campaign Quality Theory suggests entrepreneurs with low opportunity

costs are the ones driving the decrease Lower opportunity costs would suggest a willingness

to take on projects of lower quality Evidence to the contrary would undermine this

proposed mechanism

Proxy campaign quality with fundraising outcomes Market should be able to sort based on quality Four Buckets:

Unfunded, Partially Funded, Funded, Hyperfunded 0% >0%-99% 100%-199% 200%+

(1) (2) (3) (4) (5) (6) (7) (8)

DV Unfunded

Campaigns Unfunded

Campaigns Partially Funded

Campaigns Partially Funded

Campaigns Funded

Campaigns Funded

Campaigns Hyperfunded Campaigns

Hyperfunded Campaigns

Rel Time (t-6) 0.00465 -0.0334 0.280*** 0.251*** 0.133 0.125 0.0923 0.0915 (0.220) (0.238) (0.0993) (0.0889) (0.0913) (0.0937) (0.193) (0.195)

Rel Time (t-5) -0.0118 -0.0523 -0.0189 -0.0552 -0.151 -0.162 0.360 0.356 (0.173) (0.152) (0.0700) (0.0708) (0.144) (0.139) (0.222) (0.229)

Rel Time (t-4) -0.188 -0.0842 -0.0635 -0.00501 -0.0262 -0.00963 0.132 0.151 (0.118) (0.117) (0.0569) (0.0626) (0.0721) (0.0749) (0.132) (0.133)

Rel Time (t-3) 0.0168 -0.0535 0.0581 0.0166 -0.00352 -0.0139 0.183 0.176 (0.126) (0.112) (0.0657) (0.0619) (0.0548) (0.0540) (0.137) (0.138)

Rel Time (t-2) 0.0744 -0.00241 0.115*** 0.0671 -0.0493 -0.0577 -0.0314 -0.0436 (0.100) (0.0907) (0.0440) (0.0481) (0.0627) (0.0639) (0.105) (0.108)

Rel Time (t-1) Omitted Rel Time (t0) 0.0580 -0.0381 0.0819** 0.0215 -0.0411 -0.0531 -0.0718 -0.0846

(0.0855) (0.0671) (0.0342) (0.0318) (0.0397) (0.0367) (0.0628) (0.0666) Rel Time (t+1) 0.0743 -0.00370 0.0574 0.00987 -0.0343 -0.0478 0.0310 0.0192

(0.0866) (0.0773) (0.0461) (0.0460) (0.0395) (0.0410) (0.0853) (0.0889) Rel Time (t+2) -0.0296 -0.106 0.0262 -0.0209 -0.0191 -0.0336 -0.0996 -0.112

(0.0849) (0.0879) (0.0482) (0.0548) (0.0512) (0.0527) (0.106) (0.108) Rel Time (t+3) -0.0999 -0.204* -0.00556 -0.0738 -0.0448 -0.0644 -0.0535 -0.0708

(0.104) (0.114) (0.0677) (0.0799) (0.0574) (0.0588) (0.120) (0.125) Rel Time (t+4) -0.261** -0.315** -0.112 -0.141* -0.110* -0.118* -0.0455 -0.0515

(0.126) (0.139) (0.0687) (0.0753) (0.0639) (0.0659) (0.145) (0.145) Rel Time (t+5) -0.251* -0.314** -0.0664 -0.105 -0.0958 -0.104 -0.0695 -0.0804 (0.142) (0.143) (0.0727) (0.0762) (0.0718) (0.0720) (0.177) (0.178)

Year Fixed Effects Yes No Yes No Yes No Yes No

Seasonal Effects Yes No Yes No Yes No Yes No Quarter Effects No Yes No Yes No Yes No Yes N 3,444 3,444 3,612 3,612 3,549 3,549 3,171 3,171

Number of Groups 164 164 172 172 169 169 151 151

Significant Decrease Among

Unfunded Campaigns

No Significant Change Among

Hyperfunded Campaigns

Middling Decrease Among (Partially)

Funded Campaigns

Pledged Dollars It is also plausible that the platforms are targeting

downtrodden economic areas Crowdfunding capital is often local (Agrawal et al 2010)

This would indicate an excess labor pool and steadily drying capital pool for Kickstarter

Replicate our estimations with total dollars pledged as the DV A change in dollars pledged suggests the capital

available for investment is changing If dollars remain constant it suggests a shift

composition of campaigns that are launched

(1) (2)

DV Dollars

Pledged Dollars

Pledged Rel Time (t-6) 0.0521 0.0569

(0.175) (0.174) Rel Time (t-5) -0.171 -0.164

(0.191) (0.189) Rel Time (t-4) 0.112 0.103

(0.124) (0.123) Rel Time (t-3) 0.108 0.114

(0.0946) (0.0966) Rel Time (t-2) 0.0208 0.0264

(0.101) (0.102) Rel Time (t-1) Omitted Rel Time (t0) 0.0782 0.0859

(0.0586) (0.0630) Rel Time (t+1) -0.000292 0.00817

(0.0660) (0.0695) Rel Time (t+2) -0.0822 -0.0731

(0.0728) (0.0719) Rel Time (t+3) -0.0637 -0.0513

(0.0774) (0.0811) Rel Time (t+4) -0.0163 -0.0112

(0.0918) (0.0919) Rel Time (t+5) -0.0601 -0.0530 (0.109) (0.108)

Year Fixed Effects Yes No Seasonal Effects Yes No Quarter Effects No Yes N 3,612 3,612

Number of Groups 172 172

Self-Reported Profession Are individuals changing their self reported

employment after entry of the platform? Do Uber drivers report themselves as drivers?

Execute a difference in difference on iPUMS Integrated Public Use Microdata Series Current Population

Survey – largest publically available microdataset DV – Self report as a paid driver or chauffer IVs

Dichotomous Uber treatment Year, Month, and EA fixed effects

Estimators – LPM and Logit

Self-Reported Profession

(1) (2) DV Driver Driver Estimator Logit LPM Uber X 0.217*** 0.000684**

(0.0693) (0.000337) Constant -4.916*** 0.00198*** (0.0873) (0.000219) Year Fixed Effects Yes Yes Month Effects Yes Yes EA Fixed Effects Yes Yes N 1,861,144 1,657,292

Summary In this work we examine how gig-economy platforms are influencing

entrepreneurial activity Results suggest

The entry of Uber X and Postmates significantly reduces activity The effect primarily accrues among low quality projects These effects take between 9 and 15 months to manifest

Economic translation of implications Reduction of 1450 campaigns during the sample $7.5 mm in requests over a 21 month period Significantly reduces the load on crowdfunding

platforms by siphoning off low quality entrepreneurs

Contributions and Implications First glimpse into the supply side of gig-economy

platforms Novel measure of entrepreneurial activity Implications for crowdfunding platforms Insights for policy makers who are currently

debating the legality of services like Uber and Postmates

Further contribution to the growing stream of literature discussing the broader societal implications of IS

Thank You

Questions or Comments?

brad.n.greenwood@gmail.com

www.fixedeffects.com